California Opens Driverless Car Competition With Testing Regulations

smaxp (2951795) writes "California just released rules for testing autonomous vehicles on California's roads and highways. Californians will soon be seeing more autonomous vehicles than just those built by the Google X labs. These vehicles offer great promise, such as freeing the driver's attention for productivity or leisure, better safety and less congestion. It will be a while, though, before we see these vehicles on the road. From the article: 'Getting started requires the RMV’s approval of testing under controlled circumstances prior to testing on public roads. The manufactures must insure the vehicles with a $5 million surety bond. Autonomous vehicle manufacturers need a permit and test drivers need a special license. The RMV will receive applications beginning on July 1, 2014, and the permits that are granted will be announced beginning on September 1, 2014.'"

Comments

I propose a test ...

[gstoddart]

I call it the aggressive, psychotic driver who makes random, unsafe lane changes, fails to signal, and swoops across several lanes of traffic while doing well over the speed limit.

Lemme see your driverless car handle that, then we'll see.

Re:I propose a test ...

[Narcocide]

That isn't any sort of a problem for LIDAR at sufficient resolution. It remains to be seen whether it can sufficiently improve traffic flow and accident incidence/mortality rates, but personally I'm more worried about asshats who will purposefully try cloak their cars so that the automated sensors can't even see them at all, just in some misguided attempt to try to prove self-driving cars as unsafe to protect "muh freedomz!"

Don't get me wrong, I'm personally not interested in one of these self-driving contraptions, but its pretty apparent during any rush hour(s) that at least 90% of the drivers on the road couldn't beat a self-driving car's computer for accuracy to save their own lives on a good day.

Re:I propose a test ...

[Joe_Dragon]

so they can fail at makeing right turns?

Re:I propose a test ...

[ShanghaiBill]

That isn't any sort of a problem for LIDAR at sufficient resolution.

Indeed. Google regularly subjects their cars to these types of scenarios, both in simulation and on test tracks. It is odd that when people try to come up with situations that SDCs "obviously" can't handle, they so often describe situations where SDCs have a clear advantage due to their much faster reaction time.

Re:I propose a test ...

[Copid]

I think that the driver of the vehicle going the wrong way down the highway would probably be at fault.

It's easy to come up with vanishingly rare scenarios with no solution that a computer won't be able to solve (although how a human driver would do better in this situation is beyond me). Making policy based on bizarre edge cases is silly.

Re:Advanced Cruise Control

[Russ1642]

Technology needs to develop in stages. It doesn't go from concept to full-blown earth-shattering product like it does in bad science fiction movies.

Re:Why so much insurance?

[gurps_npc]

Why exactly should they? Keep in mind that when they are hit by a drunk driver they have to settle for a small insurance payout.

Your basic assumption are flawed. Specifically you appear to believe (with zero reason for these beliefs) that:

An autonomous car will be less safe than the average driver - including drunks, teenagers, old people and parents screaming at their kids to stop fighting.

That the owner should somehow be responsible in ANY way, rather than the programming company.

That the 'driver' - who in this case is not driving but instead monitoring to make sure the system is working - is in anyway responsible.

That the state would let autonomous cars drive if they were not proven better drivers than teenagers and senior citizens.

I predict that the states will have reasonable requirements for the autonomous car to prove that they are SIGNIFICANTLY better drivers than your average teenager - as in perfect driving with zero mistakes on driving tests that are far more difficult than what people have to pass.

I then predict that they will force the programming company to accept all legal responsibility for driving when the system is working - and the owner/monitor will only be required to verify that the vehicle has no warning lights activated.

I further predict that car accidents will drop to a tenth of what it is currently - and that video recordings taken by the autonomous cars will prove that except in the rare circumstances they are in an accident is because some human violated driving laws, not that the car made bad decisions

Driving is something that an artificial intelligence should be able to do far better than a human. Their reflexes are better, they don't get angry, they don't get impatient, they don't get drunk, they don't get worse as they age and they start out experienced rather than as a beginner.

Anyone that can't understand sounds like a Luddite to me.

Re:Why not?

[FlyHelicopters]

But if I have to pay attention even if under automatic control, then I don't see the point. If I have to pay attention, then I might as well do the driving myself.

Understandable reaction, but you're wrong.

Autopilots in airplanes do not remove the pilot's requirement to pay attention to what is going on. In fact, by NOT having to physically fly the plane, the pilot has a better idea of what is going on around him/her.

I can tell you from much experience that autopilots are wonderful things, you'll see more and be aware of so much more once the car does the driving.

Re:Why not?

[Obfuscant]

Autopilots in airplanes do not remove the pilot's requirement to pay attention to what is going on.

Autonomous vehicles are being promoted for exactly their ability to allow the driver to do non-driving functions (like read, eat, nap, or other things). That's what they mean when they talk about increasing productivity of those who commute to work using one.

I can tell you from much experience that autopilots are wonderful things,

Yes, they are. But they are not intended to allow flight in close formation (like a string of autos on the freeway would be), or in close proximity to the ground*. These autopilots are intended for an environment where the closest thing to you is more than 500 feet away at least. And they will quite happily fly you into the ground when they fail. Or fly you to the point you stall and then fall to the ground.

Here's just one example of how the autopilot can fail, even though the NTSB would call it pilot error (just like everything else, almost.) The Garmin G1000 with (mumble) 700 autopilot has a VS command. That's "vertical speed". You can tell the autopilot to climb at a set rate, say 500 fpm. If you forget to add power you may not be able to achieve a 500 fpm climb (or you may initially make it, but as you climb the performance decreases and you can't keep it) -- but the autopilot will keep trying. It will try to increase your angle of attack to get more lift so you can climb at the rate you've requested. It will keep trying so hard that it may cause your airspeed to drop below stall speed.

Hello, Pilot, you are now in a full-on stall, probably about to enter a spin, maybe in IMC, and your autopilot has adjusted your elevator trim to full-up trying to do what you told it to do. It's your aircraft. HAND.

It has, in it's data, the stall speeds for the aircraft it is in because it will display the critical speeds as flags on the airspeed indicator, so it could easily report the problem to the pilot. "Bong -- minimum airspeed reached, climb aborted." It does not. There is a recent article in Aviation Safety, I think it was, about a crash of a military version of a King Air in the mideast that did exactly that. The pilot was IFR and distracted and the aircraft stalled and then spun in.

There is a very good reason that there are half a dozen (8 for the G1000, as I recall) or more ways of disabling the autopilot in an aircraft. They fail often enough, and in serious enough ways ("hey, let's run your elevator trim FULL UP for no reason at all, bud", e.g.) that it is important to be able to kill George immediately. And have multiple ways to kill him in case the first three didn't kill him well enough.

You should probably not use aircraft autopilots as examples of robo-perfection.

* Yes, there are cat III autopilots that will fly the aircraft onto the runway, or "auto-land". It takes specially certified equipment AND CREWS to do that. You aren't going to find many Joe Sixpacks out on a drive that could meet equivalent quals. And that G1000 I use as an example? The aircraft manual prohibits use of the autopilot below 800 AGL, as I recall.

What I find really amazing...

[Yosho]

is how every time there's an article about autonomous cars, there are waves of people who have spent about five minutes thinking about the subject and are sure that they have come up with a laundry list of show-stopping issues that the people who've been working on this problem for a decade could not have possibly thought of.